Joint Feature Extraction and Classifier Design for ECG-Based Biometric Recognition.

Area of Science:

• Biometrics and Signal Processing
• Machine Learning in Healthcare

Background:

• Traditional biometric systems rely on physiological traits like fingerprints and faces.
• Electrocardiogram (ECG)-based biometrics are gaining traction, particularly in clinical settings.
• Existing ECG methods often separate feature extraction and classifier design.

Purpose of the Study:

• To propose a multitask learning approach for simultaneous feature extraction and classifier design in ECG biometrics.
• To improve the accuracy and efficiency of individual identification using ECG signals.

Main Methods:

• Implemented a multitask learning framework where feature extraction and classifier design occur concurrently.
• Assigned weights to features within each task's kernel.
• Decomposed feature weights into sparse (individual-specific) and low-rank (subject-common) components.
• Developed a fast, first-order optimization algorithm.

Main Results:

• The proposed approach effectively integrates feature extraction and classifier design.
• Sparse components identified individual-specific features, while low-rank components captured common feature subspaces.
• Experimental validation on the MIT-BIH Normal Sinus Rhythm database demonstrated the approach's performance.

Conclusions:

• The multitask learning method offers a unified framework for ECG-based biometric recognition.
• Simultaneous learning of features and classifiers enhances identification accuracy.
• The approach provides a computationally efficient solution for ECG biometric systems.